Mammary Carcinoma Cell Derived Cyclooxygenase 2 Suppresses Tumor Immune Surveillanceby Enhancing Intratumoral Immune Checkpoint Activity

Abstract

Introduction

Systemic inhibition of the inflammatory enzyme cyclooxygenase (COX) 2 decreasesthe risk of breast cancer and its recurrence. However, the biology of COX-2 in themulticellular tumor microenvironment is poorly defined.

Methods

Mammary tumor onset and multiplicity were examined in ErbB2 transgenic mice thatwere deficient in mammary epithelial cell COX-2 (COX-2^MECKO) comparedto wild type (WT) mice.

Tumors were analyzed, by real time PCR, immune-staining and flow cytometry, forproliferation, apoptosis, angiogenesis and immune microenvironment. LentiviralshRNA delivery was used to knock down (KD) COX-2 in ErbB2-transformed mouse breastcancer cells (COX-2KD), and growth as orthotopic tumors was examined in syngenicrecipient mice, with or without depletion of CD8⁺ immune cells.

Results

Mammary tumor onset was delayed, and multiplicity halved, in COX-2^MECKOmice compared to WT. COX-2^MECKO tumors showed decreased expression ofKi67, a proliferation marker, as well as reduced VEGFA, its receptor VEGFR2,endothelial NOS and the vascular endothelial marker CD31, indicating reduced tumorvascularization. COX-2^MECKO tumors contained more CD4⁺ Thelper (T_h) cells and CD8⁺ cytotoxic immune cells (CTL)consistent with increased immune surveillance. The ratio of T_h markersTbet (T_h1) to GATA3 (T_h2) was higher, and levels of Retnla,a M2 macrophage marker, lower, in COX-2^MECKO tumor infiltratingleukocytes compared to WT, suggesting a prevalence of pro-immune T_h1over immune suppressive T_h2 lymphocytes, and reduced macrophagepolarization to the immune suppressive M2 phenotype. Enhanced immune surveillancein COX-2^MECKO tumors was coincident with increased intratumoral CXCL9,a T cell chemoattractant, and decreased expression of T lymphocyte co-inhibitoryreceptors CTLA4 and PD-1, as well as PD-L1, the ligand for PD-1. PD-L1 was alsodecreased in IFNγ-treated COX-2KD mouse mammary cancer cells in vitro and, compared to control cells, growth of COX-2KD cells as orthotopic tumorsin immune competent mice was markedly suppressed. However, robust growth ofCOX-2KD tumor cells was evident when recipients were depleted of CD8⁺ cells.

Conclusions

The data strongly support that, in addition to its angiogenic function, tumor cellCOX-2 suppresses intratumoral cytotoxic CD8⁺ immune cell function,possibly through upregulation of immune checkpoints, thereby contributing to tumorimmune escape. COX-2 inhibition may be clinically useful to augment breast cancerimmunotherapy.

Introduction

The inducible form of cyclooxygenase (COX), COX-2, and one of its pro-inflammatoryproducts, prostaglandin (PG) E₂, are strongly implicated in a range of humancancers including breast cancer [1, 2]. Global deletion or pharmacological inhibition of COX-2 suppressedtumorigenesis in mice [3, 4] and humans [5]. PGE₂ signals through multiple pro-tumor pathways, includingPI3K/AKT, RAS-MAPK/ERK and Gs-axin-ß-catenin signaling, to increase tumor cellsurvival, inhibit apoptosis, increase cancer cell motility, stimulate angiogenesis andinhibit immune surveillance [6].

In the last decade it has become clear that the tumor microenvironment is critical fortumors to survive and progress. In addition to vascular supply, the interplay of tumorcells with non-malignant cells in the stroma provides growth, survival and motilityadvantages. A central part of the tumor microenvironment is infiltration of immunecells, which can positively or negatively influence tumor progression depending on theirdifferentiation [7, 8]. Tumor rejection is favored through T helper 1 (T_h1)-derivedcytokines that drive antigen-presenting and pro-immune M1 macrophage functions, and bythe direct tumoricidal actions of CD8⁺ cytotoxic T lymphocytes (CTLs) andnatural killer (NK) cells [6]. However, as tumors progress, soluble mediators and cellular interactions arethought to reprogram immune cells to type 2 functions so that T_h2lymphocyte-derived cytokines polarize macrophages to the M2 phenotype to suppress CTLs,promote angiogenesis and support tumor growth [6, 9]. In breast cancer, poor prognosis is associated with elevated T_h2lymphocytes and tumor-associated macrophages (TAM), while T_h1 lymphocytes,CTLs and NKs correlate with enhanced survival [8, 10], raising intense interest in therapeutic approaches to modify the tumorimmune microenvironment. COX-2-derived PGE₂ has emerged as a tumor-derivedmediator that contributes to development of immune tolerance [11–13]. Several studies report the association of tumor COX-2 with infiltrating Tcells, dendritic cells, myeloid derived suppressor cells and macrophages [14–17], while PGE₂ has been linked to immune suppression inhepatocellular carcinoma [18], lung [19], ovarian [20] and breast [14, 15] cancers. The mechanisms through which COX-2/PGE₂ suppress immunefunction are poorly defined; however, PGE₂ suppressed the ability of matureCTLs to kill murine plasmocytoma cells [21] and inhibited T_h1 generation of interferon γ (IFNγ) [22, 23], a cytokine that is critical to sustain anti-tumor immune function [6, 24].

We reported that selective deletion of mammary epithelial cell (MEC) COX-2(COX-2^MECKO) delayed carcinogen-induced mammary tumor onset coincidentwith enhanced markers of anti-tumor type 1 immunity [17]. Chemical carcinogens are generally not, however, considered significant inhuman breast cancer etiology; therefore, in the current study, we investigated the roleof tumor cell COX-2-derived mediators in ErbB2 (HER-2/neu)-induced mammarytumorigenesis. ERBB2 gene amplification or overexpression of the HER-2 protein has beenidentified in 25% to 34% of human breast cancers [25, 26]. ErBb2 mouse models show remarkable morphological resemblance to some formsof human breast cancer and accurately recapitulate the hallmark changes associated withthe early stages of human breast cancer [27]. In COX-2^MECKO mice transgenic for an activated ErbB2 mutant, wedetermined delayed tumor onset and reduced tumor multiplicity, as well as reduced tumorvascularization, compared to wild type (WT). Deletion of COX-2 in tumor cells alsosignificantly impaired maintenance of pro-tumorigenic lymphoid and myeloid cellfunctions thereby facilitating enhanced immune surveillance.

Methods

Mice and tumor tissue collection

All procedures were conducted in accordance with National Institutes of Healthregulations and were approved by the Institutional Animal Care and Use Committee ofthe University of Pennsylvania.

Floxed COX-2 mice, generated by flanking the COX-2 gene between introns 5 and 8 withloxP sites (COX-2^flox/flox), were backcrossed fully (>9 generations)onto an FVB background and are denoted as wild type (WT) mice. COX-2^flox/flox mice were crossed with FVB mice expressing Cre-recombinase under control of themouse mammary tumor virus (mmtv) promoter (Cre^mmtv), which is used widelyto target transgene expression to MEC. The resulting mice were termedCOX-2^MECKO and their characterization is described in our previous work [17]. WT and COX-2^MECKO were crossed with mice transgenic for theErbB2 (HER2/c-neu) oncogene carrying Eactivating Val⁶⁶⁴ to Glu⁶⁶⁴ mutation (Jackson Laboratory, Bar Harbor, ME, USA), also expressed under thecontrol of mmtv promoter. Genotype verification was performed by conventional PCRusing primers listed in Table 1.

Table 1 Genotyping primer sequences

Mice were palpated weekly and considered tumor bearing if a palpable mammary masspersisted for more than one week. On necropsy, tumors were counted and isolated fromsurrounding tissues, after which they were either frozen and stored at -80°C forRNA extraction or fixed in Prefer (Anatech, Battle Creek, MI, USA) overnight andparaffin embedded or digested to obtain single cell suspension for flow cytometry andmicrobead separation. For tissue digestion, tumors were washed with (D)MEM/F12 + 5%fetal bovine serum (FBS) + gentamycin 50 mg/ml, minced and placed in digestion bufferconsisting of 9 parts of wash buffer +1 part collagenase/hyaluronidase (StemCellTechnologies, Vancouver, BC Canada). After two hours shaking at 37°C, thesuspensions were centrifuged at 1,000 rpm for five minutes. Pellets were washed andtreated with red cell lysis buffer (1 part HBSS+2%FBS + 3 parts NH₄Cl) andthen with Trypsin-ethylenediaminetetraacetic acid (EDTA) 0.25% (Gibco, Grand Island,NY, USA), followed by Dispase and DNase (StemCell Technologies). Thereafter, cellpellets were passed through a 40 μm cell strainer, counted and re-suspendedeither in fluorescence-activated cell sorting (FACS) buffer for flow cytometry(description below) or in degassed MACS buffer (PBS + 0.5% BSA + 2 mM EDTA) forpositive selection of CD45⁺ cells using CD45-microbeads (Miltenyi Biotec,Auburn, CA, USA) according to the manufacturer's instructions.

NAF mammary tumor cell line culture, transduction, and treatments

The NAF tumor cell line, which was generated from mammary tumors of ErbB2-transgenicmice, was kindly provided by Dr. Lewis Chodosh (University of Pennsylvania). NAF werecultured in (D)MEM medium containing 10% FBS, 1% L-glutamine and 1%penicillin/streptomycin. For viral transduction, 15,000 cells/well were plated on96-well plates. Mission plKO.1-puro Transduction Lentiviral Particles (20 μl),carrying either non-target control small hairpin RNA (shRNA) or COX-2 shRNA (Sigma,St. Louis, MO, USA), at 1 × 10⁷ TU/ml were added to the wells with 8μg/ml protamine sulfate. After 18 hours, lentiviral particles were removed andcells kept in medium containing 2 μg/ml puromycin (Sigma) to select fortransduced cells. COX-2 knock down in COX-2 shRNA transduced NAF cells (NAF COX-2KD)compared to non-target shRNA transduced cells (NAF nt) was verified by Q-PCR. Cellswere serum starved for 24 hours and then treated with 10 ng/ml IFNγ(PeproTech,Rocky Hill, NJ, USA ), with or without 250 nM PGE₂ (Cayman Chemicals, AnnArbor, MI, USA). Fresh IFNγ and PGE₂ were added 24 hours later, andcells were harvested (0.25% Trypsin-EDTA) after 48 hour treatment, washed andre-suspended in FACS buffer for flow cytometry analysis.

Bone marrow-derived macrophage isolation and culture

Bone marrow-derived macrophages (BMDM) were isolated as described [28]. Femurs from female mice were flushed with (DMEM and cells pelleted (1,000rpm) and incubated at 37°C for 24 hours in (DMEM containing 10% FBS, 1%L-glutamine, and 1% penicillin/streptomycin. Non-adherent cells were collected andplated in L929 cell-conditioned medium (LCCM). To make LCCM, medium collected fromL929 cells (American Type Culture Collection, Manassas, VA, USA), that were split 1:5and grown to confluency, was mixed 1:5 with (DMEM with 10% FBS, 1% L-glutamine, and1% penicillin/streptomycin. Purity (approximately 99%) was verified by flow cytometryfor F4/80 and CD11b (not shown). BMDM were plated (0.5 × 10⁶ cells/well) in LCCM. At 100% confluency, media was replaced with (DMEM. After24 hours, cells received vehicle, or M1 polarizing mix (100 ng/ml lipopolysaccharide(LPS; Sigma) and 20 ng/ml IFNγ (Peprotech)), or M2 polarizing mix (20 ng/mL IL-4and 10 ng/mL IL-13; Peprotech), with or without 250 nM PGE₂ (CaymanChemicals). Supernatants were removed 18 hours later and cells lysed for RNAisolation.

Real Time RT-PCR

Total RNA from tumors and cells was isolated (RNeasy, Qiagen, Germantown, MD, USA),and reverse transcribed (TaqMan Reverse Transcriptase, Applied Biosystems, Carlsbad,CA, USA), according to the manufacturer's instructions. Real time quantitative(Q)-PCR of all genes, including 18S ribosomal RNA, was performed using inventoriedgene expression assays and TaqMan Universal PCR Master Mix from Applied Biosystems.PCR products were detected in ABI-PRISM 7900 sequence detection systems (AppliedBiosystems). Results were analyzed using the comparative Ct method, and normalized to18S RNA.

Immunohistochemistry

Paraffin embedded tumor tissues were sectioned (4 μm). After de-paraffinizationand rehydration, endogenous peroxidase was blocked with 3% hydrogen peroxide. Heatinduced epitope retrieval was performed with 1 mM EDTA (Invitrogen, Grand Island, NY,USA). After overnight blocking at 4°C with 5% donkey serum (Sigma) + 0.1% Triton100× (Sigma), sections were incubated with primary antibodies overnight at4°C as follows: anti-Ki67 (Abcam, Cambridge, MA, USA,1:50 dilution), anti-CD31(Abcam, 1:200 dilution) or anti-CXCL9 (Aviva Systems Biology, San Diego, CA, USA1:125 dilution). Thereafter, the Polink-2 HRP Plus AEC System forImmunohistochemistry (Golden Bridge International, Inc, Mukillteo, WA, USA) was used,according to the manufacturer's instructions. Slides were then counterstained withhematoxylin (Vector Laboratories, Burlingame, CA, USA), and mounted (Aqua-Mount;Lerner Laboratories, Pittsburgh, PA, USA). Images were taken with a Nikon EclipseE600 microscope using ACT-1 imaging program (Nikon Instruments Inc., Hicksville, NY,USA).

Flow cytometry

Single cell suspensions from tumor digestions (above) were centrifuged, washed andre-suspended in FACS buffer (PBS + 2% FBS +1 mM EDTA + 0.01% sodium azide), 1 ×10⁶ cell/100 μl/tube. After a five minute incubation with ratanti-mouse CD16/CD32 (Mouse BD Fc Block, BD Pharmingen, Franklin Lakes, NJ, USA), 1μg/ml of fluorescein isothiocyanate (FITC) conjugated anti-CD3, PE conjugatedanti-CD4, and AF647 conjugated anti-CD8a, or PE conjugated anti-F4/80 and AF647conjugated anti-CD86 antibodies (Invitrogen) were added. OneComp eBeads (eBioscience,San Diego, CA, USA) were incubated with anti-CD3 or anti-CD4 or anti-CD8a antibodiesto perform compensation for spectral overlap. NAF COX-2KD and NAF nt (1 ×10⁶ cell/100 μl), were incubated with PE conjugated anti-PD-L1antibody (Biolegend, San Diego, CA USA). After a 30-minute incubation, cells werewashed and re-suspended in 500 μl FACS buffer. Unstained tumor cells and cellsincubated with isotype control rat anti-mouse antibody were used as negativecontrols. FACS analysis was performed on a BD FACSCalibur machine (BD Biosciences,San Jose, CA, USA). Data was analyzed using FlowJo Research Flow Cytometry AnalysisSoftware (TreeStar, Ashland, OR USA).

Orthotopic tumor growth and CD8⁺depletion

NAF COX-2KD and NAF nt tumor cells were injected into the #4 and #9 mammary glands (1× 10⁶ cells/gland in 100 μl Hanks Balanced Salt Solution) ofnormal WT female mice between 8 to 14 weeks of age. Orthotopic tumor volume wasdetermined weekly using standard caliper measurement. For CD8+ depletion experiments,mice were injected intraperitoneally with 200 μg of an anti-CD8 or isotypecontrol antibody (BioXCell, West Lebanon, NH, USA), four days and again two daysprior to injection of tumor cells, and then twice weekly for a further four weeks.Depletion of CD8+ cells was confirmed by flow cytometry of erythrocyte lysed wholeblood (ACK Lysing Buffer, Invitrogen), four days and again four weeks after tumorcell injections.

Statistical analysis

Statistical analyses were performed using Prism (GraphPad Software, Inc., La Jolla,CA, USA). As appropriate, comparisons were made using logrank analysis, unpairedt-test (with Welch's correction when variances were significantly different byF-test), Mann Whitney test (when the data distribution was not normal), or, formultiple group comparisons, analysis of variance (ANOVA) followed by Bonferroni'smultiple comparison test.

Results

Tumor onset, development, and vascularization in WT and COX-2 ^MECKOmice

The current investigation was designed to study the role of MEC COX-2 in mammarytumor development, with the goal of elucidating whether and how targeted inhibitionof COX-2 in epithelial cells affects the disease. In our previous study we confirmedCOX-2 deletion in MEC isolated from COX-2^MECKO mice by Q-PCR and Westernblotting, and loss of PGE₂ generation by COX-2^MECKO cells wasestablished by mass spectrometry [17]. COX-2 expression and PGE₂ production were unchanged inperipheral macrophages isolated from COX-2^MECKO compared to WT confirmingthe selectivity of the deletion [17]. In the current study, tumor onset was significantly delayed inCOX-2^MECKO mice compared to their WT littermates (Figure 1A). On necropsy, COX-2^MECKO mice had significantly fewertumors compared to WT (Figure 1A). Consistent with theseobservations, cell proliferation appeared higher in WT tumors, as indicated by higherlevels of mRNA for the proliferation marker Ki67 in WT compared toCOX-2^MECKO tumors (Figure 1B). Markers forapoptosis (caspase3) and autophagy (Lc3) were not different between the two genotypes(Figure 1B). Abundant expression of Ki67 protein in WT, but notCOX-2^MECKO, tumors was confirmed by immunohistochemistry (Figure 1C).

Figure 1

Tumor onset, multiplicity and cell proliferation were suppressed inCOX-2^MECKO tumors. COX-2^MECKO tumors aredenoted as KO. (A) Percent of tumor free mice against weeks of age. Meantumor free time for COX-2^MECKO mice was 29 weeks versus 23 weeks forWT (left graph, n = 13 to 19). The right graph shows tumormultiplicity as number of tumors per mice at necroscopy (n = 14 to18). (B) Gene expression levels of Ki67 (proliferation), Caspase3(apoptosis) and Lc3 (autophagy) in whole tumors by Q-PCR (n = 8 to18). (C) Immunohistochemistry staining for Ki67 (dark red-brown) insections of paraffin embedded WT and COX-2^MECKO tumors (image shownis representative of n = 4). Cell nuclei are counterstained withhematoxylin. The bar on the KO panel indicates 20× magnification. Data incolumn graphs are mean ± sem. P values are compared to WT; *P < 0.05. COX, cyclooxygenase; KO, knock out; WT, wild type.

Q-PCR analysis of tumors revealed lower expression levels of CD31, an endothelialmarker, endothelial (e) NOS, the angiogenic factor VEGFA and its receptor VEGFR2(Figure 2A), in COX-2^MECKO compared to WT. Althoughno difference was observed in mRNA levels of the lymphangiogenic factor VEGFC, itsreceptor VEGFR3 was significantly lower in COX-2^MECKO tumors (Figure 2A). Immunostaining for CD31 revealed a denser blood vesselnetwork in WT tumors, confirming suppressed angiogenesis in COX-2^MECKOtumors (Figure 2B).

Figure 2

Angiogenesis was suppressed in COX-2^MECKO tumors.COX-2^MECKO tumors are denoted as KO. (A) Gene expressionlevels for CD31, eNOS, VEGFA, VEGFC, VEGFR2 and VEGFR3 in whole tumors byQ-PCR. Data are mean ± sem of n = 8 to 18. P values arecompared to WT. *P < 0.05. (B) Immunohistochemistry stainingfor CD31 (dark red-brown) in sections of paraffin embedded WT andCOX-2^MECKO tumors (image is representative of n = 6).Cell nuclei are counterstained with hematoxylin. The bar on the KO panelindicates 20× magnification. COX, cyclooxygenase; KO, knock out; WT, wildtype.

Subpopulations and phenotypes of tumor infiltrating immune cells in WT andCOX-2^MECKO tumors

WT and COX-2^MECKO tumors were analyzed by flow cytometry and Q-PCR tocompare the populations of infiltrating immune cells and their phenotypes. By flowcytometry, there was no difference in the total number of F4/80⁺ TAMsbetween WT and COX-2^MECKO tumors (Figure 3A).COX-2^MECKO tumors did, however, have significantly higher numbers ofCD3⁺CD4⁺ cells, a population that includes T_h1,T_h2, and regulatory T (T_reg) cells, as well asCD3⁺CD8⁺ CTLs and CD3^-CD8⁺ cells,encompassing NK and dendritic cells (Figure 3A). To furtherdefine their functional identify, tumor-infiltrating leukocytes (TILs) were isolatedusing magnetic microbeads coated with a pan-leukocyte marker CD45 and cells analyzedby Q-PCR for phenotypic markers and cytokines (Figure 3B). Theratio of Tbet (T_h1 marker)/GATA3 (T_h2 marker) tended to behigher in COX-2^MECKO tumors compared to WT (Figure 3B), suggesting a prevalence of CD3⁺CD4⁺ T_h1 over T_h2 lymphocytes, when MEC COX-2-derivedmediators are absent. Further, mRNA levels for either Tbet alone or the Tbet/GATA3ratio were significantly correlated with CD4 mRNA in COX-2^MECKO, but notWT, tumors (data not shown). Gene expression of FoxP3, a marker for T_reg,was not altered and there was no difference in mRNA for macrophage type 1 cytokinesTNFα and IFNγ or an M1 macrophage marker CD86, in CD45⁺ TILs,suggesting no major change in M1 polarization in this disease model. COX-2-derivedPGE₂ has been implicated in driving the immune suppressive phenotypetypically associated with TAM [6]. Indeed, exogenous PGE₂ treatment significantly increased theexpression of M2 marker Arginase 1, a key enzyme in suppression of T cell function,in both M1 and M2 polarized bone-marrow derived macrophages (Figure 3C). In tumors, although arginase 1 mRNA levels were similar in TILs fromCOX-2^MECKO and WT tumors, another M2 marker, Retnla, was significantlydecreased in COX-2^MECKO (Figure 3B). Taken as awhole, our flow cytometry, immune staining and CD45⁺ cell expressionanalysis indicates that absence of epithelial COX-2-derived mediators augmentsT_h1 and cytotoxic immune function and reduces immune suppressivemacrophage function in the mammary tumor microenvironment.

Figure 3

Deletion of mammary epithelial COX-2 modified tumor infiltrating cellphenotype. COX-2^MECKO tumors are denoted as KO. (A) Flow cytometry analysis of tumor associated lymphocytes and macrophages(n = 7 to 10). (B) Gene expression levels measured by Q-PCRin CD45⁺ tumor infiltrating leukocytes (TILs) isolated from wholetumors by microbead separation (n = 6 to 7). (C) Geneexpression levels measured by Q-PCR in unpolarized, M1 polarized or M2polarized bone marrow derived macrophages, treated without or with PGE₂ (n = 8). Data are mean ± sem. P values arecompared to WT; *p < 0.05. COX, cyclooxygenase; KO, knock out;PGE₂, prostaglandin E₂; WT, wild type.

COX-2 may enhance immune tolerance through suppression of T cell recruitment andactivation

Our data thus far indicates a significant contribution of mammary epithelialCOX-2-derived mediators to pro-tumor immune function, particularly T lymphocyte andcytotoxic immune cell function, in the tumor microenvironment. We next examinedpathways that control T cell recruitment, activation and function. In breast cancer,tumor cell expression of the chemokines CXCL9 and 10 recruits lymphocytes, improvessurvival in mouse models and human studies [29, 30], and PGE₂ inhibits expression of both chemokines in breastcancer cells in vitro [12]. Paraffin embedded sections of WT and COX-2^MECKO tumors showedsubstantially higher levels of CXCL9 expression, by immunohistochemistry, inCOX-2^MECKO tumors, and this staining was evident throughout the tumorcells (Figure 4A). WT tumors, in contrast, showed weak CXCL9staining (Figure 4A). T cell activation requires binding of Tcell receptors to antigen and is regulated by a balance of co-stimulatory andco-inhibitory receptor-ligand interactions. T cell CD28 receptor engagement by CD80or CD86, expressed on antigen presenting cells, provides the additional signalnecessary for T cell activation. The same ligands can, alternatively, drive T cellsto a state of anergy through binding to cytotoxic T lymphocyte antigen 4 (CTLA-4) [31]. Inhibition of T cell function is also directed through binding ofprogrammed death ligand 1 (PD-L1) to its receptor, PD-1, expressed on the T cellsurface [32]. In our study, gene expression levels for both inhibitory receptors CTLA4and PD-1, as well as PD-L1, were decreased in COX-2^MECKO tumors comparedto WT, suggesting suppressed signaling through co-inhibitory pathways (Figure 4B). Both cancer cells and tumor infiltrating myeloid cells areconsidered as sources of PD-L1 expression in the tumor microenvironment [33, 34]. We did not observe any change in PD-L1 mRNA levels in CD45⁺ TILs from COX-2^MECKO and WT tumors (data not shown), suggesting thattumor cell PD-L1 was suppressed by COX-2 deficiency. Indeed, NAF COX-2KD, which,compared to NAF nt, grew poorly as orthotopic tumors in immune competent syngenicmice (Figure 5B) also produced substantially less PD-L1 proteinin response to IFNγ (Figure 5C). Interestingly, additionof exogenous PGE₂ neither modified PD-L1 expression in NAF nt nor rescuedIFNγ-induced PD-L1 expression in NAF COX-2KD cells.

Figure 4

COX-2 suppresses immune surveillance by altering T cell recruitment andactivation. COX-2^MECKO tumors are denoted as KO. (A) Immunohistochemistry staining for CXCL9 (brown) of sections from paraffinembedded WT and COX-2^MECKO tumors (two images from each genotypeshown are representative of n = 4). Nuclei are counterstained withhematoxylin. The bar on the KO panel corresponds to 40× magnification.(B) Gene expression levels measured by Q-PCR in whole tumors (n = 8 to 18). Data are mean ± sem. P values are compared toWT; *P < 0.05. COX, cyclooxygenase; KO, knock out; WT, wildtype.

Figure 5

COX-2 knock down reduced tumor cell PD-L1 expression and orthotopic tumorgrowth. ErbB2-transformed mammary epithelial cells (NAF) were transducedwith non-targeted shRNA (NAF nt) or COX-2 targeted shRNA (NAF COX-2KD). (A) COX-2 mRNA levels measured by Q-PCR, after 24 hours of treatment with 5μg/ml LPS (n = 4). (B) Percent of tumor free mice aftermammary fat pad injection of NAF nt and NAF COX-2KD (n = 6). (C) Mean fluorescence of PD-L1 in NAF, measured by flow cytometry, after 48hours of treatment with 10 ng/ml IFNγ, with or without 250 nM PGE₂ (n = 3 to 6). Data in column graphs are mean ± sem.P values are compared to NAF nt control, unless otherwiseindicated. *P < 0.05, **P < 0.005, ***P <0.0005. COX, cyclooxygenase; KO, knock out; shRNA, short hairpin RNA.

To assess how critical the loss of COX-2's immune suppressive actions was for reducedtumor growth and burden, we examined growth of NAF COX-2KD orthotopic tumors inrecipient mice treated with an anti-CD8 antibody, to deplete CD8⁺ immunecells, or an isotype control antibody. Complete depletion of CD8⁺ cells inblood was confirmed by flow cytometry (Figure 6A). In isotypecontrol antibody treated mice NAF COX-2KD grew poorly in only two of six injections.In contrast, six of six NAF COX-2KD tumors grew in CD8⁺ depleted mice,similar to NAF nt control cells (Figure 6B), and were markedlylarger at necroscopy (four weeks after tumor injection; Figure 6C).

Figure 6

CD8⁺ cell depletion restores growth of injected NAF COX-2KD tumorcells as orthotopic tumors. (A) Flow cytometry of peripheral blood fromanti-CD8 or isotype control antibody treated mice four days after mammary fatpad injection (representative of n = 6). (B) Tumor growth of NAF ntand NAF COX-2KD cells injected into mammary fat pads of mice that receivedeither anti-CD8 or isotype control antibody injection (antibody treatmentdenoted with ^ along X axis; n = 6). (C) NAF COX-2KD tumor volumemeasured on necropsy of animals that were treated with either isotype controlor anti-CD8 antibody. In B and C data are mean ± sem of n = 6.**P < 0.005, ***P < 0.0005. COX, cyclooxygenase.

Discussion

Significant attention is now focused on understanding how resident and infiltratingcells in the tumor microenvironment support disease progression and in developingtherapeutic strategies directed at microenvironmental targets [7]. Central to the pro-tumor microenvironment is suppression of immune cellfunction allowing tumor cells to avoid destruction. In the current study, wedemonstrated enhanced immune cell recruitment and reduced T cell co-inhibitory pathwaysin tumors that lack mammary epithelial expression of the pro-inflammatory enzyme COX-2,coincident with delayed ErbB2 oncogene-driven mammary tumor development.

Consistent with established paradigms of COX-2 in cancer [6, 35, 36], deletion of MEC COX-2 delayed mammary tumor onset, lowered tumormultiplicity, reduced tumor cell proliferation and decreased tumor vascularization.Reduced vascularization in COX-2^MECKO tumors was associated with lowerexpression of VEGFA and its receptor VEGFR2, a dominant pro-angiogenic pathway in tumors [37], consistent with the role of COX-2 in promoting the angiogenic switch thatallows tumors to progress [38]. It may be that reduced tumor cell proliferation and suppressed angiogenesisassociated with deletion of MEC COX-2 was sufficient to suppress tumors. However, theelevation of CD4⁺ and CD8⁺ immune cell populations we observed inCOX-2^MECKO mice, prompted us to consider how tumor cell COX-2 contributesto tumor immune function.

COX-2-mediated promotion of pro-tumorigenic T_h2 lymphocyte and M2 macrophagefunctional phenotypes, as well as suppression of cytotoxic immune cell activity, hasbeen reported [6]. However, it remains unclear how COX-2 contributes to the orchestration ofimmune cell function as tumors develop. In part, the paucity of information reflects thedifficulties of working with global COX-2 knock out mice, which have breeding problems,severe renal pathology and a shortened life span [39], none of which are encountered in our targeted COX-2^MECKO mice, aswell as the extensive use of immune deficient host mice for tumor transplant studies.Compared to WT, three populations of immune cells - CD3⁺CD4⁺,which are T_h lymphocytes, CD3⁺CD8⁺ cells, which areCTLs and CD3^-CD8⁺, which encompass NKs and dendritic cells - wereelevated in COX-2^MECKO tumors. Within the CD3⁺CD4⁺ population, an increase in anti-tumorigenic T_h1 cells may suppresstumors in COX-2^MECKO mice; however, greater activity of T_h2lymphocytes and/or T_reg would be expected to promote tumor growth [24]. The strong trend towards an increased T-bet/GATA3 mRNA ratio, a measure ofthe T_h1 to T_h2 balance [40], and the unchanged expression of FoxP3, a marker for T_reg [6, 24], indicates the likely prevalence of the pro-immune helper function ofT_h1 lymphocytes over pro-tumorigenic T_h2 lymphocytes or immunesuppressive T_regs, in COX-2^MECKO tumors. These data are consistentwith the shift toward type 1 immunity we reported previously in carcinogen-inducedmammary tumors in COX-2^MECKO mice, which were also delayed compared to WT [17].

Within the CD8⁺ populations, the suppressed tumor phenotype inCOX-2^MECKO mice may result from increased cytolytic actions of CTLs andNKs [24], as well as enhanced immunogenic actions of mature dentritic cells [41]. We did not directly discriminate between the relative contributions of theseCD8⁺ subtypes; however, a key role for CD8⁺ immune cells inCOX-2-mediated control of tumor immune function is strongly supported by the restorationof NAF COX-2KD tumor cell growth in CD8⁺-depleted mice.

TAM are abundant in mammary tumors and their density is generally directly correlatedwith disease severity and prognosis [42, 43]. Similar to the T_h1 and T_h2 lymphocytecharacterization, M1 and M2 macrophages are considered anti- and pro- tumor,respectively [44]. We reported previously that COX-2-derived PGE₂ restrains M1macrophage polarization in vitro and in carcinogen-induced mammary tumors [17]. In the current model, however, CD86, a M1 macrophage marker, was notdifferent in COX-2^MECKO tumor associated F4/80⁺ cells(macrophages) or in isolated CD45⁺ TILs, compared to WT. It is likely thatthe relevance of COX-2-mediated paracrine control of M1 macrophage function to tumorprogression varies between models. Retnla (Resistin-like molecule alpha/FIZZ1), acytokine derived from alternatively activated M2 type macrophages [45], was significantly lower in CD45⁺ TILs from COX-2^MECKOtumors suggesting reduced M2 polarization, a possible reflection of reducedT_h2-derived cytokines in the COX-2^MECKO microenvironment and/orloss of paracrine COX-2-derived PGE₂ activity, which augments M2 polarizationof BMDM in vitro.

As a whole, our analysis of the tumor microenvironment strongly supported a shifttowards enhanced helper and effector T lymphocyte recruitment and function inCOX-2^MECKO tumors. It may be that there is simply an increased immune cellrecruitment to breast tumors lacking epithelial COX-2. Indeed, we saw a dramaticincrease in tumor cell expression of the T cell chemokine CXCL9 in COX-2^MECKOtumors, consistent with a recent report in patients with invasive breast cancer thattumor cells are the major source of CXCL9 [12]. In the same study, PGE₂ suppressed IFNγ-induced CXCL9 levelsin MCF-7 and MDA-MB 231 breast cancer cells, and COX inhibitors increased CXCL9secretion. Despite the higher CXCL9 levels in COX-2^MECKO tumors, however, theabsolute number of CD3⁺ cells by flow cytometry was not higher than in WTtumors suggesting a local influence of tumor cell COX-2 derived mediators in limitingimmune cell function rather than a simple recruitment effect.

Intense interest in cancer immunotherapy has focused recently on immune checkpoints,whose function to dampen immune responses is important for self tolerance and control ofphysiological immune responses. Two central and well-studied immune checkpoints are theco-inhibitory receptors CTLA4 and PD-1; antagonists to both are currently in clinicaltrials for melanoma and other cancers [32]. Engagement of CLTA4 or PD-1 on immune cells by their ligands CD80/CD86 orPD-L1, respectively, can suppress or shut down immune surveillance [46]. Conversely, blockade of co-inhibitory receptor-ligand interaction canenhance anti-tumor immunity [32]. In our study, levels of CTLA4 and PD-1, as well as PD-L1, were decreased inCOX-2^MECKO tumors. The PD-1-PD-L1 interaction is of particular interest inthis regard since PD-1 expression in tissues is induced by inflammatory signals where itacts to suppress T cell activity and limit collateral tissue damage [32]. We reasoned, therefore, that COX-2, an established inflammatory gene, mayact in tumors to upregulate expression of PD-1/PD-L1, thereby suppressing immunefunction and facilitating immune escape. In support of this hypothesis, NAF COX-2KD,which grew very poorly as orthotopic tumors, generated substantially less PD-L1 inresponse to IFNγ compared to NAF nt control cells. The failure of exogenousPGE₂ to restore PD-L1 expression levels in NAF COX-2KD may suggestdistinct actions of autocrine and paracrine PGE₂, or indicate a role forother COX-2-derived products, in tumor cell COX-2 mediated control of PD-1 expression.The pathways through which COX-2-derived PGE₂/other prostanoids control tumorcell expression of PD-L1 and other immune modulators are currently underinvestigation.

Our study provides significant insight into the complex autocrine and paracrinefunctions of mammary epithelial COX-2 in ErbB2-induced breast cancer and suggests thattumor cell COX-2 is an important component in establishing a permissive immunemicroenvironment. Recent studies indicated that CD8⁺ tumor infiltrationbolstered chemotherapeutic responses in human breast cancer and mouse models [8]. Our demonstration that deletion of tumor cell COX-2 can enhancetumor-associated CD8⁺ cytotoxic immune cell infiltration and function mayopen new avenues to develop targeted strategies for COX-2 inhibition in combination withcytotoxic drugs. Further, there have been significant advances in cancer immunotherapyusing antibodies to block CTLA4 or PD-1 co-inhibitory function, thereby augmentinganti-tumor immunity [32, 47–49]. To our knowledge, our study is the first to link COX-2 to T cellco-inhibitory receptor/ligand function, a potentially new avenue to investigate COX-2inhibitors as adjuvants to immunotherapy. Finally, we demonstrated that interruption ofCOX-2 function selectively in epithelial cells was sufficient to reduce ErbB2- (thisstudy) and carcinogen [17] induced mammary tumorigenesis and growth. The clinical use of systemic COX-2inhibitors in cancer, although supported across multiple studies [1], is limited by the associated gastrointestinal and cardiovascular hazards [50]. We speculate that, as improved targeted drug delivery modalities continue toemerge, delivery of COX-2 selective inhibitors directly to the tumor cells may allow forsafe and effective use of these drugs in cancer without the deleterious side effectsassociated with systemic COX inhibition.

Conclusions

The data strongly support that, in addition to its angiogenic function, tumor cellCOX-2-derived mediators suppress anti-tumor immune cell function, possibly throughupregulation of inhibitory immune checkpoints, contributing to tumor immune escape.COX-2 inhibition may be clinically useful to augment breast cancer immunotherapy.